Friday, October 20, 2017

Over-Estimating Value of NHL Corsi

We may have reached a point in the NHL community where Corsi (or shot differential +/-) is being over-used in analyzing performance. If you take all the players who played in both 2015/16 and 2016/17, there is only a 0.25 (or 25%) correlation between Corsi For % from one season to the next. How many shots on goal an individual player gets per game is 80% correlated year over year, +/- is at 20%. There is a troubling amount of randomness in individual Corsi. Shot differential while a player is on the ice is more a function of the other players around him and the sum of the situations he's placed in by the coaches.

Think about it this way, for most of a game a player is 1 of 12 on the ice. How much of an impact does each really have on the shot totals? I believe there are only a small number of players on each team who actually drive possession, and everyone else is just along for the ride making a minimal contribution to the shot clock. It's probably something like 25% of the players have a meaningful impact on shot differential, the other parts are interchangeable.

One group frequently that frequently gets overrated by Corsi is bottom pairing defenseman, where defensive defensemen who play some of the most difficult minutes will often have a poor Corsi For. The bottom pairing defensemen get to play their ice time against the bottom forward lines from the other team, where the defensive guy will often play against the best opposing forwards. A common roster mistake made by teams who place extra emphasis on shot differential is to replace the low Corsi guys playing tougher minutes with a high Corsi guy playing easier minutes. Then when the "analytics darling" gets put into tougher situations, he can't handle it.


I do think that there can be some usefulness to shot differential statistics, if viewed in the proper context, comparing players who play in similar situations. When I see that Drew Doughty or Patrice Bergeron has a consistently high Corsi For, that makes sense. The two most important Corsi For percentages on your roster should be the #1 Defenseman and the #1 Center. I do care how those players affect the shots on goal battle. But if you're going to use Corsi to promote or demote players up and down the line-up, be warned that it's a slippery slope. 

No comments:

Post a Comment